microsoft/PEACE

GeoMap-Bench是一个为了评估多模态语言模型在地质图理解方面的性能的基准数据集。它包含了来自USGS（英文）和CGS（中文）的图像-问题对及注解答案，共有124张图像和3864个问题。这些问题分为三种类型：选择题、填空题和论述题。数据集覆盖了提取、定位、指代、推理和分析五种能力，并定义了25个任务。

GeoMap-Bench is a benchmark dataset designed to evaluate the performance of multimodal language models on geologic map understanding. It consists of image-question pairs with annotated answers from USGS (English) and CGS (Chinese), totaling 124 images and 3,864 questions. The questions are categorized into three types: multiple-choice, fill-in-the-blank, and essay. The dataset covers five abilities: extracting, grounding, referring, reasoning, and analyzing, and defines 25 tasks.